kaboot/server/src/services/documentParser.ts
2026-01-23 15:23:54 -07:00

346 lines
9.7 KiB
TypeScript

import officeParser from 'officeparser';
import WordExtractor from 'word-extractor';
import * as XLSX from 'xlsx';
import { execSync } from 'child_process';
import { writeFileSync, readFileSync, unlinkSync, existsSync } from 'fs';
import { tmpdir } from 'os';
import { join } from 'path';
import { randomUUID } from 'crypto';
export const GEMINI_NATIVE_TYPES = [
'application/pdf',
'text/plain',
'text/markdown',
'text/csv',
'text/html',
'text/css',
'text/javascript',
'image/jpeg',
'image/png',
'image/gif',
'image/webp',
'image/heic',
'image/heif',
];
export const TYPES_TO_CONVERT_TO_PLAIN_TEXT = [
'text/xml',
'application/xml',
'application/json',
'application/javascript',
'text/x-python',
'text/x-java-source',
'text/x-c',
'text/x-c++',
'text/x-typescript',
'text/yaml',
'application/x-yaml',
];
// MIME types that officeparser can extract text from (modern Office formats)
export const OFFICEPARSER_TYPES = [
'application/vnd.openxmlformats-officedocument.wordprocessingml.document', // .docx
'application/vnd.openxmlformats-officedocument.presentationml.presentation', // .pptx
'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet', // .xlsx
'application/vnd.oasis.opendocument.text', // .odt
'application/vnd.oasis.opendocument.presentation', // .odp
'application/vnd.oasis.opendocument.spreadsheet', // .ods
'application/rtf', // .rtf
];
// Legacy Office formats requiring specialized parsers
export const LEGACY_DOC_TYPE = 'application/msword'; // .doc (word-extractor)
export const LEGACY_XLS_TYPE = 'application/vnd.ms-excel'; // .xls (xlsx/SheetJS)
export const LEGACY_PPT_TYPE = 'application/vnd.ms-powerpoint'; // .ppt (ppt/SheetJS)
// Image types that can use OCR for text extraction
export const OCR_CAPABLE_TYPES = [
'image/jpeg',
'image/png',
'image/gif',
'image/webp'
];
export const SUPPORTED_TYPES = [
...GEMINI_NATIVE_TYPES,
...OFFICEPARSER_TYPES,
...TYPES_TO_CONVERT_TO_PLAIN_TEXT,
LEGACY_DOC_TYPE,
LEGACY_XLS_TYPE,
LEGACY_PPT_TYPE,
];
const EXTENSION_TO_MIME: Record<string, string> = {
'.txt': 'text/plain',
'.md': 'text/markdown',
'.markdown': 'text/markdown',
'.csv': 'text/csv',
'.html': 'text/html',
'.htm': 'text/html',
'.xml': 'application/xml',
'.json': 'application/json',
'.js': 'text/javascript',
'.mjs': 'text/javascript',
'.ts': 'text/x-typescript',
'.tsx': 'text/x-typescript',
'.jsx': 'text/javascript',
'.py': 'text/x-python',
'.java': 'text/x-java-source',
'.c': 'text/x-c',
'.h': 'text/x-c',
'.cpp': 'text/x-c++',
'.hpp': 'text/x-c++',
'.cc': 'text/x-c++',
'.css': 'text/css',
'.yaml': 'text/yaml',
'.yml': 'text/yaml',
};
export function normalizeMimeType(mimeType: string, filename?: string): string {
if (SUPPORTED_TYPES.includes(mimeType)) {
return mimeType;
}
if (mimeType === 'application/octet-stream' && filename) {
const ext = filename.toLowerCase().match(/\.[^.]+$/)?.[0];
if (ext && EXTENSION_TO_MIME[ext]) {
return EXTENSION_TO_MIME[ext];
}
}
if (mimeType.startsWith('text/') && !SUPPORTED_TYPES.includes(mimeType)) {
return 'text/plain';
}
return mimeType;
}
export interface ProcessedDocument {
type: 'native' | 'text';
content: Buffer | string;
mimeType: string;
}
export interface ProcessOptions {
useOcr?: boolean;
}
export function isSupportedType(mimeType: string): boolean {
return SUPPORTED_TYPES.includes(mimeType);
}
export function isGeminiNative(mimeType: string): boolean {
return GEMINI_NATIVE_TYPES.includes(mimeType);
}
export function isOcrCapable(mimeType: string): boolean {
return OCR_CAPABLE_TYPES.includes(mimeType);
}
export function needsOfficeParser(mimeType: string): boolean {
return OFFICEPARSER_TYPES.includes(mimeType);
}
export function needsPlainTextConversion(mimeType: string): boolean {
return TYPES_TO_CONVERT_TO_PLAIN_TEXT.includes(mimeType);
}
export function needsWordExtractor(mimeType: string): boolean {
return mimeType === LEGACY_DOC_TYPE;
}
export function needsSheetJS(mimeType: string): boolean {
return mimeType === LEGACY_XLS_TYPE;
}
export function needsSheetJSPPT(mimeType: string): boolean {
return mimeType === LEGACY_PPT_TYPE;
}
async function extractWithOfficeParser(buffer: Buffer, useOcr: boolean = false): Promise<string> {
const config = useOcr ? {
extractAttachments: true,
ocr: true,
ocrLanguage: 'eng'
} : {};
const ast = await officeParser.parseOffice(buffer, config);
let text = ast.toText();
if (useOcr && ast.attachments) {
for (const attachment of ast.attachments) {
if (attachment.ocrText) {
text += '\n' + attachment.ocrText;
}
}
}
return text;
}
async function extractWithWordExtractor(buffer: Buffer): Promise<string> {
const extractor = new WordExtractor();
const doc = await extractor.extract(buffer);
return doc.getBody();
}
function extractWithSheetJS(buffer: Buffer): string {
const workbook = XLSX.read(buffer, { type: 'buffer' });
const textParts: string[] = [];
for (const sheetName of workbook.SheetNames) {
const sheet = workbook.Sheets[sheetName];
const csv = XLSX.utils.sheet_to_csv(sheet);
if (csv.trim()) {
textParts.push(`--- ${sheetName} ---\n${csv}`);
}
}
return textParts.join('\n\n');
}
// Map legacy extensions to their modern equivalents for LibreOffice conversion
const LEGACY_TO_MODERN: Record<string, string> = {
'.ppt': 'pptx',
'.doc': 'docx',
'.xls': 'xlsx',
};
async function extractWithLibreOffice(buffer: Buffer, extension: string, useOcr: boolean = false): Promise<string> {
// Use LibreOffice to convert legacy Office files to modern format, then parse
const tempId = randomUUID();
const tempDir = tmpdir();
const inputPath = join(tempDir, `input-${tempId}${extension}`);
const modernExt = LEGACY_TO_MODERN[extension] || 'pdf';
const outputPath = join(tempDir, `input-${tempId}.${modernExt}`);
try {
// Write input file
writeFileSync(inputPath, buffer);
// Convert to modern format using LibreOffice
try {
execSync(
`libreoffice --headless --convert-to ${modernExt} --outdir "${tempDir}" "${inputPath}"`,
{ timeout: 60000, stdio: 'pipe' }
);
} catch (execError) {
// LibreOffice not available or conversion failed
const error = execError as Error & { code?: string };
if (error.code === 'ENOENT' || error.message?.includes('not found')) {
throw new Error(
`Legacy ${extension} files require LibreOffice for text extraction. ` +
`Please convert to .${modernExt} format or ensure LibreOffice is installed.`
);
}
throw new Error(`LibreOffice conversion failed: ${error.message}`);
}
// Read the converted file and extract text using officeparser (with OCR if enabled)
if (existsSync(outputPath)) {
const convertedBuffer = readFileSync(outputPath);
const config = useOcr ? {
extractAttachments: true,
ocr: true,
ocrLanguage: 'eng'
} : {};
const ast = await officeParser.parseOffice(convertedBuffer, config);
let text = ast.toText();
// Include OCR text from attachments if available
if (useOcr && ast.attachments) {
for (const attachment of ast.attachments) {
if (attachment.ocrText) {
text += '\n' + attachment.ocrText;
}
}
}
return text;
}
throw new Error('LibreOffice conversion produced no output');
} finally {
// Cleanup temp files
try { unlinkSync(inputPath); } catch { /* ignore */ }
try { unlinkSync(outputPath); } catch { /* ignore */ }
}
}
export async function processDocument(
buffer: Buffer,
mimeType: string,
options: ProcessOptions = {}
): Promise<ProcessedDocument> {
if (!isSupportedType(mimeType)) {
throw new Error(`Unsupported file type: ${mimeType}`);
}
// Images with OCR requested - extract text
if (isOcrCapable(mimeType) && options.useOcr) {
const text = await extractWithOfficeParser(buffer, true);
return {
type: 'text',
content: text,
mimeType: 'text/plain'
};
}
// Gemini-native types (including images without OCR) - pass through
if (isGeminiNative(mimeType)) {
return {
type: 'native',
content: buffer,
mimeType
};
}
// Office documents - extract text, OCR extracts text from embedded images
if (needsOfficeParser(mimeType)) {
const text = await extractWithOfficeParser(buffer, options.useOcr);
return {
type: 'text',
content: text,
mimeType: 'text/plain'
};
}
// Legacy .doc files (Word 97-2003)
if (needsWordExtractor(mimeType)) {
const text = await extractWithWordExtractor(buffer);
return {
type: 'text',
content: text,
mimeType: 'text/plain'
};
}
// Legacy .xls files (Excel 97-2003)
if (needsSheetJS(mimeType)) {
const text = extractWithSheetJS(buffer);
return {
type: 'text',
content: text,
mimeType: 'text/plain'
};
}
// Legacy .ppt files (PowerPoint 97-2003) - use LibreOffice for conversion
if (needsSheetJSPPT(mimeType)) {
const text = await extractWithLibreOffice(buffer, '.ppt', options.useOcr);
return {
type: 'text',
content: text,
mimeType: 'text/plain'
};
}
if (needsPlainTextConversion(mimeType)) {
return {
type: 'text',
content: buffer.toString('utf-8'),
mimeType: 'text/plain'
};
}
throw new Error(`No extraction handler for: ${mimeType}`);
}